Essential Elements of PHI: Predictions

Predictions are among the most essential elements of PHI investment decisions, planning and evaluation. Yet they are the least likely elements to be present in the typical application. This may be due to general skepticism over the accuracy and reliability of predictions, but the effects of failure to generate needed predictions can be worse than the limitations of the predictions themselves.

Perhaps the most important prediction needed in PHI planning and evaluation is the depiction of the future without intervention. While the most common approach seems to be to rely on the most recent year’s experience as baseline for planning and evaluation, the best approach is to use a prediction for the next year if no PHI intervention is applied. When dealing with health, diseases, and risks, it is the normal situation for next year to be worse than this year, absent any intervention.

Predicting the next year realities of the metrics to be used in evaluating PHI is far better than relying on after vs. before comparisons. Moreover, when applied to disease management, after vs. before comparisons are likely to ignore the natural tendency of healthcare costs, for example, to “regress to the mean”, i.e. people with outlier levels of high expenditures in the baseline year are far more likely to have lower expenditures in the following year as a matter of course, whether or not any PHI intervention is applied.

Failure to take into account this regression to the mean tendency will result in significant over-counting of reductions in expenditures as consequences of interventions, rather than recognizing them as natural phenomena. A simple after vs. before comparison will ignore a host of factors that will tend to drive expenditures higher, such as inflation, effects of population aging, and worsening of risks without interventions.

But the most serious and most common problem is the failure to predict the likelihood of success, for individuals and populations. Literally thousands of studies have demonstrated that only a portion of populations enroll in PHI efforts, make the desired health behavior changes, achieve desired health status improvements and increases in productivity or performance. The probability and accuracy of predictions of such effects determines the success of PHI investments, so are essential in making good investment decisions.

Yet the vast majority of predictions rely on predicted costs of inaction, alone, rather than the probable savings of action. If the predicted costs without intervention are as high as $5000 per employee or plan member, for example, but only 30% of the population will enroll, only 50% of those will make the desired behavior change, 50% of those achieve the desired health status improvement, and 50% of those demonstrate improved productivity/performance, at an average gain in value of 5%, then the predicted effect would be only $5000 x .30 x .50 x .50 x .50 x .05 = $9.375 per member of the population.

If the costs of intervention amount to as little as $10 per population member, the results would be a net financial loss, though 7.5% of the population would gain improved health (30% enrollment x 50% making health behavior change x 50% achieving improved health = 7.5%). Without significantly higher levels of success at some point(s) in the cascade of value, there would be little reason to invest.

By contrast, an accurate prediction of the probability of success can add greatly to both the overall investment decision and the selection of targets for intervention, as well as the matching of interventions to individuals. If the predicted value of individuals or segments can be made and is reasonably accurate, interventions can be assigned to each based on the combination of potential (e.g. predicted healthcare costs and productivity/performance levels without intervention) x probability of realizing it.

This combination would at least enable the “scaling” of interventions to the predicted value of individuals or segments. Sponsors could use a predetermined level of ROI to make such matches, e.g. if they insist on a 2:1 ROI, they would set the limit for the cost of interventions to ½ the predicted combination of value. They could even use the same process to empower participants to select any combination of intervention costs and incentives they can reward themselves with for achieving the predicted results.

In any case, creating and continuously improving predictions of potential value gains times the probability of achieving them -- for populations, segments, or individuals -- should greatly improve the strategic and tactical investment decisions that sponsors can make. By contrast, relying solely on baseline measures and potential gains is as likely to turn out badly as well, depending on what the probability of success really is, and how well matches of investment are made.

There has long been an unfortunate practice in quoting an old saying as “Ignorance is bliss; ‘tis folly to be wise.” The accurate quotation is “If ignorance is bliss…etc.”. In the case of planning and making PHI investment decisions, ignorance of the potential/probability and value of success is not merely unwise, it could be disastrous.